blog / 11/12/2026 / 4 min read

Menu chatbot or AI agent: which should you choose?

A menu works for closed questions. An agent works when customers write their own way and send several messages in a row.

Choose a menu chatbot if your customers ask closed questions that fit into a few buttons. Choose an AI agent if customers write their own way and send several messages in a row, because in that case the menu only pushes the conversation to a person on the team.

What each one does well

A menu chatbot offers numbered options and follows a fixed path. For opening hours, address or a copy of a payment slip, that works. The question is closed and the answer does not change.

An AI agent reads what the customer wrote and replies based on the business’s information and rules. It does not require the customer to pick an option before speaking.

The difference shows up when the conversation does not follow the script. Imagine a customer who does not type “2”. They write “hi”, then ask their question, then send “can I pay in installments?”, all in a row. A menu does not know what to do with that.

The cost the menu hides

In a menu, a question that does not fit a button ends at the same exit: talk to a person on the team. In the flow diagram that looks solved. In practice, someone has to be on the other side to finish the conversation.

At the businesses I serve, before the agent, the person who answered WhatsApp was the owner or one employee. That is a general answer, not a measurement. On a team that size, every conversation the menu hands off lands on one of those people, and they have working hours.

And customers do not write only during working hours. At a car audio shop I serve, 38.9% of customer messages arrived outside the shop’s hours: 8,816 of 22,650, between January 19 and August 18, 2026. The hours used in the count are the hours rule in the shop’s registration, and I was the one who applied that rule to the data. Outside those hours, the button to talk to someone leads to a person who is off the clock.

How the agent handles a loose conversation

In the agent I maintain for that shop, a message goes through these steps:

  1. 01Receivesthe customer's message arrives on WhatsApp
  2. 02Groupsconsecutive messages from the same customer
  3. 03Repliesbased on the shop's catalog and rules
  4. 04Simulatesinstallments in 3, 6 and 10 payments
  5. 05Hands offpasses the conversation to a person when the topic goes beyond what it knows

The grouping step is the one that handles a customer who writes in pieces. The agent replies to the whole, not to each loose piece.

Look at the last step: the agent also hands conversations to a person. The difference is the point where that happens. In a menu, the handoff happens when the question does not fit a button. In the agent, it happens when the topic goes beyond what it knows.

Between October 11 and November 10, 2026, this agent sent 5,516 replies. Of those, 2,108 went out outside the shop’s hours.

  • 5,516replies from the agent
  • 2,108replies outside the shop's hours
A car audio shop, October 11 to November 10, 2026

From April to July 2026, the median response time stayed between 3 and 6 seconds, inside and outside business hours. In hours, the 5,516 replies come to about 91.9 h of typing, on an estimate of 1 minute per reply. That premise has no measured basis and only gives an order of magnitude.

When a menu is enough

The agent is not the answer for every case. It needs work after it is built: adjusting rules, switching models, finding the cause of an incident. I do that work on the agents I maintain. If a menu solves your business’s questions, that extra work is not justified.

In my opinion, the test is simple. If the question has a fixed answer and the customer is fine picking an option, the menu handles it. If the answer depends on what the customer explains, and they explain it across several messages, the menu will hand that conversation off.

What these numbers do not show

The numbers above measure service: a message that arrived and a reply that went out. They do not measure sales, because the data I collect from the agent has no sales record. I also have no data from before the agent to compare with. What I can state is coverage of hours.

Run the test on your WhatsApp

List the ten questions your customers send most. For each one, ask: can it be answered with a button, without the customer having to explain anything?

If almost all of them fit a button, start with a menu. If several depend on details the customer has to explain, also count how many arrive outside your hours. Those are the conversations that today wait for someone on the team.

To see how the car audio shop’s agent works, the full case is at WhatsApp agent for an automotive shop.

Have a process that could be automated?

Tell me how it works today. I’ll say whether it can be done, how long it takes and what changes.

Message me on WhatsAppthe first conversation is free